Yuan Yao, 北京大学

Title

Some Open Problems in Online Algorithms

Abstract

In this talk, I would like to discuss some interesting yet open problems arising from online algorithms in ranking and learning. The first problem is to obtain error bounds in exponential probabilistic inequalities, which lead to almost-sure convergence as well as optimal in both convergence rates and condition number in constants. The second problem is related to stochastic approximations of sparsity regularization paths, e.g. stochastic linearized Bregman iterations with early stopping.


« Back